Text data labeling
Deepen AI's trained teams classify text and tag entities in your categories, for training data or for your operations, and send you the hard cases.
From $5 per person-hour A plan within 2 working days No sales call
What a labeled item looks like
Each item comes back labeled in your categories, with the reason for every change to a pre-label.
- Text
- My card was charged twice for the same order.
- Pre-labeled
Billing: refund request- Labeled as
- CorrectedBilling: duplicate charge
- Reason
- Your categories separate duplicate charges from refund requests.
- Text
- Ship 40 units of SKU A-1192 to the Leeds warehouse by Friday.
- Pre-labeled
- Quantity: 40. Product: A-1192. Location: Leeds.
- Labeled as
- CompletedDate: Friday, added. The other three are correct.
- Reason
- Your spec tags relative dates as well as exact ones.
- Text
- Great, another update that resets all my settings.
- Pre-labeled
Positive- Labeled as
- CorrectedNegative
- Reason
- “Great” is sarcastic here. Your guidelines label by meaning, not by individual words.
- Text
- Can I bring my assistance dog to the Oxford store?
- Pre-labeled
- Store policy
- Labeled as
- Sent to youIt could be “Store policy” or “Accessibility”. Which should assistance animals go under?
- Reason
- Your answer is added to the category definitions for everyone.
How we start on your work
No sales call. A plan in your inbox, then a short paid pilot on your real text.
What we need from you
- Your categories or taxonomy, with a short definition for each
- 50 to 100 items you've already labeled
- Access to the text: your labeling tool, Ground Truth or a shared sheet
- 5 minutesTell us the task
What text, how many items, and what the labels are for.
- Within 2 working daysGet a plan and a price
People, hours, turnaround, start date and the total price.
- 2 weeksRun the pilot
We train the team on your categories, label real text and send two weekly reports. Then you decide.
Who labels, and how labels stay consistent
Trained people, working in English. Each one learns your categories and passes a test on known items before live work.
Every reviewer's accuracy is measured
Items you've already answered are hidden in the real work, so we know how accurate each person is.
Nobody guesses on unclear cases
The team lead decides, or asks you. Your answer joins the guidelines.
You see the results every week
A weekly report shows work done, QA results and what's improving.
Standard labeling covers support tickets, reviews, messages, survey answers, search queries and documents in English. Other languages and specialist text are quoted in your plan.
How much does text data labeling cost?
$5per person-hour
Standard labeling in English. You pay for the hours people work, with a trained team, a team lead, QA checks and a weekly report included.
Set before you pay. Your plan shows the team size, hours and total price.
Starts with a two-week pilot on your real text, paid upfront, with two weekly reports.
Your data stays in your systems. We work under NDA with the access you set. Deepen AI holds SOC 2 Type II, ISO 27001, ISO 9001 and TISAX.
Questions
Which labeling tools can the team use?
Yours, SageMaker Ground Truth as a private workforce, or a shared sheet. Your plan says which.
Can an item have more than one label?
Yes. Your guidelines set whether items take one label or several, and which labels can go together.
How do you keep labels consistent?
Items with known labels are mixed into the work, a share of items is labeled by more than one person to check agreement, and QA re-checks a sample. The weekly report shows the results.
What happens when our categories change?
Tell your team lead. The team retrains on the change, and the weekly report notes the date, so you know which labels followed which version.
Is this for training data or for our operations?
Either. The same team can label text to train a model, or sort live tickets and messages so they reach the right place.
What format do labels come in?
Your schema, in your tool, a sheet, or JSON or CSV files, with the item ID, the labels and any notes.